Walmart Sales Forecasting

Machine learning project for predicting weekly sales using advanced time series analysis, feature engineering, and XGBoost regression on Walmart historical data.

🎯 Overview

End-to-end sales forecasting pipeline with comprehensive EDA, anomaly detection, multiple feature selection techniques, and XGBoost time series modeling for Walmart's 400k+ sales records across 45 stores.

⚡ Key Features

• Exploratory Data Analysis (EDA): Statistical insights, correlation analysis, and data visualization
• Anomaly/Outlier Detection: Statistical and IQR-based outlier identification and removal
• Advanced Feature Engineering: Time-based, lag, rolling window, Trend and seasonality, and interaction features (130+ features)
• Multiple Feature Selection: Univariate, Mutual Information, XGboost Feature importance
• Data Quality Pipeline: Comprehensive cleaning, validation, and problematic value handling

📊 Dataset

Walmart historical sales data (2010-2012):

• 400,000+ sales records across 45 stores and multiple departments
• Store metadata, economic indicators (CPI, unemployment, fuel prices)
• Holiday and seasonal information

🔧 Implementation

Pipeline

1. EDA & Visualization → 2. Anomaly Detection → 3. Feature Engineering → 4. Feature Selection → 5. XGBoost Modeling

Feature Selection Methods

• Univariate statistical testing
• Mutual Information scoring
• XGboost importance ranking

Model

• XGBoost Regressor: Optimized for time series forecasting
• Time-aware feature engineering for temporal patterns
• Cross-validation and hyperparameter tuning

📈 Results

• 130+ engineered features from raw data
• Robust outlier detection improving data quality
• Multi-algorithm feature selection optimizing model performance
• XGBoost time series model for accurate sales forecasting